MiniMax H3 Spicy API

MiniMax H3 open weights, self-hosted and unfiltered. Text, image, or reference to video at 480p or 720p, with turbo / standard / high compute tiers and synced audio on every one. Integrate this model via REST with endpoint docs, parameters, and code examples.

Prompting guide Compare video models Use cases

Video Generation

Img2Vid

Audio

Content moderation disabled

Credit Cost

10 credits

Type

Video

Get API Key

Create auth keys for apps and agents

See Usage

Track requests, credits, and logs

Buy Credits

Top up prepaid API capacity

API Endpoint

POST

/api/v1/models/minimax-h3-spicy/run

Test This Model

Run a real request through the API pipeline and inspect the exact input and output your integration will handle. Uses your credits like any API call; failed runs are refunded automatically.

LLM.txt

AI-Ready

For AI assistants

Copy this model's documentation into Claude, GPT, or any AI for instant integration help.

/llm/minimax-h3-spicy.txt

OpenAPI Spec

Swagger compatible

Import into Postman, Swagger UI, or generate client SDKs in any language.

/api/openapi

Parameters

Input parameters accepted by this model

Parameter	Type	Description	Default
`prompt` required	string	Text description of the video, motion, and sound to generate	-
`image_url`	url	First-frame image. Switches to image-to-video mode.	-
`end_image_url`	url	Last-frame image. Requires image_url.	-
`reference_images`	string	Up to 9 image URLs that guide subject and scene identity. Runs the Ref2VA checkpoint. Cannot be combined with image_url / end_image_url.	-
`reference_videos`	string	Up to 3 reference clips (2-15s, 24fps) that guide motion and style. Each clip's own soundtrack is used as an audio reference unless reference_video_audio is false. Runs the Ref2VA checkpoint.	-
`reference_audios`	string	Up to 3 standalone audio references — a voice, a music bed, or an ambience the generated soundtrack should follow. Runs the Ref2VA checkpoint.	-
`reference_video_audio`	boolean	Whether each reference video's own soundtrack is passed as an audio reference. Default true.	`true`
`quality`	enum	Compute tier, multiplies the resolution rate: turbo is fastest at 0.5x price (step-distilled, softer motion and quieter audio), standard is 1x, high is 1.5x more sampling for the best fast-motion detail at 2x price. `turbostandardhigh`	`standard`
`resolution`	enum	Output resolution. Sets the base rate: 480p is 1 credit per second, 720p is 2. `480p720p`	`720p`
`aspect_ratio`	enum	Output aspect ratio. Ignored in image-to-video mode — the start frame sets the aspect. `16:94:31:19:163:4`+1 more	`16:9`
`duration`	integer	Video duration in seconds (5-15) min: 5 \| max: 15	`5`
`seed`	integer	Random seed. Omit and one is chosen for you; the response reports which seed it used. Reusing a seed with identical settings reproduces the same video (GPU kernels are not bit-deterministic, so it is visually identical rather than byte-identical). Reusing it at a different quality tier changes the sampling trajectory, so you get the same idea and framing rather than the same clip.	-

Supported Aspect Ratios

16:9

4:3

1:1

9:16

3:4

21:9

Code Examples

Text → video — copy and paste to get started

bash

curl -X POST "https://pixeldojo.ai/api/v1/models/minimax-h3-spicy/run" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A lone rider crosses a storm-lit desert ridge, thunder rolling",
    "duration": 5,
    "aspect_ratio": "16:9"
  }'

Response Format

Submit Response (202 Accepted)

json

{
  "jobId": "job_abc123...",
  "status": "pending",
  "statusUrl": "https://pixeldojo.ai/api/v1/jobs/job_abc123...",
  "creditCost": 10,
  "creditsRemaining": 95,
  "expiresAt": "2026-08-08T19:40:00Z"
}

Completed Response

json

{
  "jobId": "job_abc123...",
  "status": "completed",
  "output": {
    "video": "https://temp.pixeldojo.ai/pixeldojotemp/1705930123-456789.mp4"
  },
  "creditCost": 10,
  "expiresAt": "2026-08-08T19:40:00Z"
}

LLM Documentation Try in Web UI Get API Key Usage Dashboard Buy Credits